It's just not possible to create a generic tool to convert a complex composite audio file to midi. Think about it...It would have to recognise instruments from wave patterns, know the exact characterisitics in order to extract each instrument, know how to deal with voices and understand interaction between instruments. You would also have to know the acoustic characteristics of the recording, as well as a host of other "featurettes".
At the moment, developers are struggling to craete true instrument synthesis, because no instrument sound can be described mathematically by one statement. The various elements change along the musical scale. So if they can't be truly synthesised, you won't be able to extract them.
I just Googled and found an article that says music recognition accuracy for a single instrument is 60% to 70%.
This is quite a good quote to get your head around the concept...
Quote: "There are 50 musicians on stage...They all start playing a different piece of music at the same time. Do you have any idea how that sounds? It's still only those 50 musicians you liked so much before, but do you think you could make a MIDI file out of it? "
BatVink
